by Raph Levien
3 Oct 2001
While reading a bit on the current W3C patent controversy in Linux Weekly News, I came across a
posting describing a patent taken out by Apple on alpha compositing.
Looking at patent US5379129 itself, I see that it contains no new intellectual content. Usually, when I see a claim that Company X has patented Obvious Technique Y, I'm skeptical. Whether you agree with the patent system or not, in many cases the actual patent contains claims that show that Company X has at least used Obvious Technique Y in a new context, or figured out an implementation trick.
Here, however, the patent really is as dumb as it looks at first glance. Every person whose name appears on it should be deeply ashamed of themselves - the inventors, Konstantin Othmer (now a venture capitalist) and Bruce A. Leak (later a cofounder of WebTV), for being so unaware of the field that they actually believed it patentable, the Apple lawyer, Mark Aaker, for not bothering to find out whether it was patentable, and of course the primary examiner, Bentsu Ro, for letting the patent go through. And Mr. Ro has been honored on the Patents Recognition Wall, at that!
At the time that this patent was filed in 1992, the technique of integral alpha had already been published for about 12 years: Smith, A.R., Painting Tutorial Notes, SIGGRAPH '79 course on Computer Animation Techniques, 1979, ACM. Alvy Ray Smith went on to receive an Academy Award for this work in 1996. In the interim, this seminal work has been cited by many, many other researchers, including Bruce Wallace's 1981 SIGGRAPH paper, and the landmark Porter-Duff paper, which added the additional innovations of premultiplied alpha and an image algebra:
Porter, T. and T. Duff, Compositing Digital Images. Computer Graphics, 1984. 18(3): p. 253-259.
However, the Apple patent contains no trace of these later innovations. Everything in it is in the Wallace paper's description of Alvy Ray Smith's work, and in essentially identical mathematical terminology to boot.
Thus, this patent can easily serve as a poster child for well known techniques, widely published in the academic literature, and then patented.
Apple is kindly agreeing to provide the use of this patent on "reasonable and non-discriminatory" terms for use in the SVG standard for vector graphics. Needless to say, the lawyers at Apple who are involved in this fiasco should also be ashamed of themselves.
Update 2001 Oct 11: Apple has posted an amazingly strong anti-patent position in response to the W3C patent controversy.
Update 2001 Nov 20: I missed something important. The main focus of this patent is per-channel alpha compositing, ie having a separate alpha channel for each of the color channels. That's not present in the Porter-Duff work. Claim 1 still suffers from the ridiculous overbroadness described above. Per-channel transparency is described in the literature (including Foley, van Dam, Feiner, and Hughes), but is just a little more work to track down.
Also please see this call for prior art on the PNG/MNG tools website.